Published by Forbes.com on January 19, 2024
On Oct. 18, 2023, I attended and participated in the Leading With AI Responsibly conference hosted by my institution, the Institute for Experiential AI. I came out with many insights from this meeting, in which business leaders shared several in-depth stories about how they’re using generative AI (GenAI) to drive business success today.
The discussions reinforced my long-held belief that a successful AI strategy requires a few specific principles for businesses that want to gain a competitive advantage with AI while avoiding its many pitfalls. Below I share seven keys for making generative AI work for businesses.
1. Human oversight brings success.
The effective use of AI, especially in business, hinges on human involvement. If humans aren’t providing oversight and feedback, models can quickly go off the rails for even the most seemingly simple applications. “Experiential AI” is the term we use for AI with a human in the loop, and we see confirmation of this in every working AI deployment out there.
An example is using large language models (LLMs) to expedite the creation of FAQ content on your website while using human reviewers to ensure the content is accurate and of high quality before publishing. Such an approach allows you to leverage the speed and productivity gains of AI while avoiding the pitfalls of using AI irresponsibly. This is especially important in highly regulated industries like finance and health care, but the risks of using AI irresponsibly extend to almost every industry. Responsible AI solutions that work with and learn from human intelligence are a better path to building working solutions that avoid the pitfalls of autonomous AI.
2. Create a sandbox.
Executives also need a thoughtful and controlled approach to implementing generative AI. A crucial tenet of this strategy is the creation of a “sandbox” environment, akin to a laboratory, where developers can experiment with and refine AI applications. This approach ensures that AI development is done responsibly and safely with a strong emphasis on data protection and ethical considerations.
For instance, companies could create internal software environments that offer developers a controlled space to experiment with AI applications. This approach encourages innovation without creating new risks. Many companies target the output of these AI models for use by their employees to evaluate fit and correctness and do not allow such applications to deal with external parties directly, thus limiting risks while learning from the sandboxed efforts.
3. Keep applications narrow.
One unfortunate byproduct of the endless AI hype is that some people believe they can build one AI model that does everything for their business. That approach simply doesn’t work and leads to unnecessary waste of resources and time. AI applications should instead be focused and narrow.
It’s important to clearly define AI use cases early on in model development. This approach minimizes potential risks and guides decisions on model techniques, data requirements, and legal or ethical issues. The lesson here is to keep your applications as narrow and as precisely defined as possible—general AI is the enemy of today’s successful deployments!
4. Data and context are key.
We all know that most working AI is highly dependent on learning algorithms (which include generative AI algorithms), and it is obvious that ML and data science require data. The data captured has to be fine-grained and detailed (for algorithmic consumption) and an important but often missed part of data collection is capturing context (technical, business and otherwise). For instance, why was a certain action or decision appropriate given the context?
Be sure to capture data with machine consumption in mind rather than just human insights, and also capture as much of the context as possible.
5. Business context is also important.
Everyone focuses on the models, but just as important for AI success is how a business is structured around those models. Is there proper oversight? Are there workflows in place to take advantage of the insights gained from AI and halt any unintended outputs of models? Can you measure if your AI models are actually helping your business improve the customer experience or the bottom line?
At the Institute for Experiential AI, we advocate for creating organizations with an internal, proactive management team or tapping into experts who can help business leaders on demand. In operational settings, proving viability, sustainability and ROI is just as critical as avoiding risks.
6. Don’t forget about predictive AI.
While generative AI and LLMs have captured attention, the reality is that not all problems can be addressed by GenAI. Many problems are best addressed with predictive AI, which constitutes the majority of working AI applications out there and has been worked on for the past four decades.
Predictive AI can often produce faster, more stable and more reliable solutions much more economically. While GenAI will help us address many tasks in accelerating knowledge economy tasks, predictive AI continues to have a huge role to play.
7. Using AI responsibly brings real business value.
I am more convinced than ever that proper and responsible use of AI tools can bring real business value. Companies are now using AI to automate tedious tasks, supercharge efficiency, and improve the accuracy and effectiveness of several products and processes.
In this rapidly evolving economic landscape, I believe that companies that successfully and responsibly harness AI’s capabilities, while also acknowledging and addressing its limitations, are poised to outpace those that remain hesitant to adopt these technologies.
View online